Skip to main content
Gervin Kahunde
In use

Admissions Application Monitoring

A high-traffic admissions application produces plenty of data. A weekly usability report turns it into decisions, so problems surface from evidence instead of support calls.

  • Product Operations
  • Research and analytics
  • Quality assurance

AI-assisted step Human decision Observed friction

Summary

Challenge
Give the product team a reliable, weekly view of where applicants struggle in the ASU admissions application, and why.
Role
For a couple of months I ran the team's weekly usability-monitoring report: pulling the metrics, reviewing session recordings, investigating anomalies, and writing the findings, before handing it to a teammate. I also QA new builds and contributed to the dashboard redesign.
Team
Product manager (my lead), engineering, UX design, and the analytics team
Duration
2026, weekly for a couple of months
Methods
Funnel analytics, session-recording review, anomaly investigation, QA, dashboard design
Outcome
Weekly reports delivered on schedule and refined with my manager's feedback, then handed to a teammate. Internal figures aren't published here.

Context and problem

The admissions application is often a prospective student's first real interaction with the university, and it serves a very large number of people across online and in-person programs. Small problems at that scale — a confusing label, an intermittent server error — quietly cost applicants.

The data to spot them existed across analytics, dashboards, and session recordings. The weekly report is where someone looks at all of it on a regular rhythm and turns it into plain-language findings the team can act on.

My role and collaborators

What I did
Ran the weekly usability-monitoring report end to end for a couple of months: pulling metrics, reviewing session recordings, investigating anything unusual, and writing and sharing the findings. I also QA new application builds and helped redesign the application dashboard.
What the team owns
My product manager sets priorities and reviews the report; engineering and design own fixes; the analytics team owns the underlying dashboards and data sources.

Constraints

  • Applicant data is personal and protected, so reporting stays aggregate.
  • Data lives in several tools that don't connect directly, and they can disagree because of different definitions and refresh delays.
  • At this volume, some errors are real but too rare to trigger alerts — the report has to separate noise from patterns.
  • Readers are busy and cross-functional, so the report has to be quick to scan.

What the report covers

Starts
Applications started, split by online and in-person, because they behave differently.
Funnel
Page-by-page completion, to show exactly where applicants stop.
Errors
Error rate and server errors — including how many applicants leave right after one.
Duration
Median time to complete, where long times suggest confusion.
Help
Help clicks by page, which point to the questions applicants find hardest.
Sessions
A sample of session recordings, focused on rage clicks, dead clicks, and errors.

Evidence and insights

Numbers say where something changed; recordings say why. Pairing them is what made the report useful. A few examples of the kind of findings it surfaced:

  • A cluster of server timeouts on one verification step — too few to trigger an alert, but worth flagging and watching.
  • Whether applicants completed a new identity-verification step, came back afterwards, or tried to get around it.
  • A sign-in label that implied some options were only for new users, when they worked for everyone.
  • A tracking tag firing twice under slightly different names, which made one metric look far smaller than it was.
  • Two analytics tools reporting different counts for the same page, with cookie-consent settings as a likely cause.

Decisions and tradeoffs

Last seven days, not 30
I standardized every metric on a seven-day lookback so the report matches its weekly cadence, even where dashboards default to 30 days.
Change, not just totals
After feedback from my manager, I added week-over-week change and made funnel drops explicit, so readers see what moved instead of re-reading the same numbers.
Watch enough sessions, not all
About ten recordings a week is enough when things look normal; when something looks off, I keep watching until the pattern is clear.
Look for the absence of bad experiences
Daily movement is mostly noise. The job is to notice when something deviates from the baseline and investigate that.

Operating model

  1. Step 01

    Collect signals

    Funnel, errors, duration, help clicks

  2. Step 02

    Compare

    Against the prior seven days and known releases

  3. Step 03 Human decision

    Watch sessions

    Focus on rage clicks, dead clicks, errors

  4. Step 04 Friction

    Investigate

    Size it and find the likely cause

  5. Step 05

    Report

    Plain-language weekly findings

  6. Step 06 Human decision

    Team decides

    Product and engineering act on it

The weekly monitoring loop.

QA and dashboards

  • Compared new application builds against production and documented visual regressions — inconsistent icon sizes, mismatched styles, misaligned elements, and a missing section.
  • Tested every sign-in method for both new and returning applicants, and diagnosed a tester problem as browser-specific so the team could move on.
  • Worked with the analytics team on a redesign of the application dashboard: clearer layout and filters, tables in place of hover-only tooltips, and realistic sample data so stakeholders react to the design rather than placeholder numbers.
  • Researched what Tableau could and couldn't do, then proposed one job per tool: Tableau for the performance overview, Looker Studio for satisfaction, errors and data-layer insights, FullStory for funnel retention and time on page, and Google Sheets for verbatim feedback. The goal was the "so what" in one view, without mental math. The redesign is currently paused.

Running it week to week

The report arrived on a predictable day with a predictable structure. I sent drafts early so my manager had time to review, and adjusted the template as the team's questions changed. Some weeks it was a ten-minute read; others needed more — the findings decided, not the template. That predictability is also what let me hand it off to a teammate without losing the rhythm.

Outcomes

While I ran it, the team got a consistent weekly view of the application, and issues were raised from evidence rather than discovered through support calls. My manager named me as the person product managers can delegate testing, user flows, documentation, and data and measurement work to.

What I learned

Monitoring is a communication product. The analysis matters, but cadence, format, and honesty about confidence decide whether anyone acts on it. And recordings are where the numbers start to make sense.

What I'd improve next

  • Spend more structured time on full session recordings and help clicks to find where applicants need the most help.
  • Track each recommendation through to its effect, so the report shows which findings led to changes.

Continue with work that uses a similar approach.

Research

Tuition Calculator Post-Launch Research

After a redesigned tuition calculator launched, many visitors left before seeing a cost — and the analytics alone couldn't explain why.

My role
Reviewed session recordings, logged behavior, and drafted the product recommendations
Evidence
Combined session recordings, analytics, and survey responses into evidence-based recommendations for the calculator team.
  • Research and analytics
  • Strategy